counties <- left_join(
read_csv("data/county_data.csv"),
read_csv("data/county_population.csv"),
join_by(GEOID == fips)
)R Bootcamp: Day 3 AM
Data visualization: base R and ggplot2
Yufei Qin
Eric Manning
September 1, 2026
Your first plot
plot(counties$income, counties$rent)
The gist: (data w/ mapping) + geom + …
ggplot(counties, aes(x = income, y = rent)) +
geom_point()
(Yuck.)
Aesthetics
Map a third column to color. It’s a variable, so it goes in aes() (constants go in geom_*)
ggplot(counties, aes(x = income, y = rent, color = region)) +
geom_point(size = 0.7)
Compare: in base R, you subset the data, chose the colors, and drew the legend.
Classic mistake
ggplot(counties, aes(x = income, y = rent, color = "steelblue")) +
geom_point(size = 0.7)
Inside aes(), "steelblue" isn’t a color. It gets vectorized as data.
Same mapping, different geom
Distributions by group:
ggplot(counties, aes(x = region, y = income)) +
geom_boxplot()
Composition: stacked bars
Split each region’s counties into high- and low-population halves:
counties <- counties |>
mutate(pop_level = if_else(
pop >= median(counties$pop, na.rm = T),
"high pop", "low_pop"
))
counties |>
filter(!is.na(pop_level)) |>
## piping in this data; now different operator!
ggplot(aes(x = region, fill = pop_level)) +
geom_bar()
Counts vs. shares: position
position = "fill" rescales every bar to 1
counties |>
filter(!is.na(pop_level)) |>
ggplot(aes(x = region, fill = pop_level)) +
geom_bar(position = "fill")
Side by side: position = "dodge"
counties |>
filter(!is.na(pop_level)) |>
ggplot(aes(x = region, fill = pop_level)) +
geom_bar(position = "dodge")
Three versions of the same bars — stacked, filled, dodged. Which question does each answer best?
Line graphs
ggplot2 ships with practice data: economics is monthly US macro data:
ggplot(economics, aes(x = date, y = unemploy)) +
geom_line()
Common mistake
head(economics_long, 3)# A tibble: 3 × 4
date variable value value01
<date> <chr> <dbl> <dbl>
1 1967-07-01 pce 507. 0
2 1967-08-01 pce 510. 0.000265
3 1967-09-01 pce 516. 0.000762ggplot(economics_long, aes(x = date, y = value01)) +
geom_line()
Tell ggplot what belongs together
Map the series id to color (or group) and each variable gets its own line:
ggplot(economics_long, aes(x = date, y = value01, color = variable)) +
geom_line()
Reference lines are geoms too
geom_hline(), geom_vline(), and geom_abline() are just more layers:
ggplot(counties, aes(x = income, y = rent)) +
geom_point(size = 0.5, color = "gray") +
geom_hline(
yintercept = median(counties$rent, na.rm = T),
color = "steelblue", linewidth = 1
)
Facets
One line splits the plot by a categorical column:
ggplot(counties, aes(x = income, y = rent)) +
geom_point(size = 0.4) +
facet_wrap(~region) ## more than one: ~ region + ...
Facets in two directions
facet_grid(rows ~ cols):
counties |>
filter(!is.na(pop_level)) |>
ggplot(aes(x = income, y = rent)) +
geom_point(size = 0.4) +
facet_grid(pop_level ~ region)
Labels with labs()
ggplot(counties, aes(x = income, y = rent, color = region)) +
geom_point(size = 0.7) +
labs(
title = "Rent rises with income",
subtitle = "US counties, 2023 five-year ACS",
x = "Median household income ($)",
y = "Median gross rent ($)",
color = "Region"
)
Scales: how mappings become positions and colors
Every aesthetic has a scale named scale_<aesthetic>_<flavor>() with its own arguments:
ggplot(counties, aes(x = income, y = rent)) +
geom_point(size = 0.5) +
scale_x_continuous(
name = "Median household income",
breaks = c(50000, 100000, 150000),
labels = c("$50k", "$100k", "$150k"),
limits = c(0, 200000)
)
Zooming vs. dropping
xlim()/ylim()/lims() throw away data outside the limits:
ggplot(counties, aes(x = income, y = rent)) +
geom_point(size = 0.5) +
lims(x = c(0, 100000))Warning: Removed 158 rows containing missing values or values outside the scale range
(`geom_point()`).
Zooming vs. dropping
coord_cartesian() zooms without dropping:
ggplot(counties, aes(x = income, y = rent)) +
geom_point(size = 0.5) +
coord_cartesian(xlim = c(0, 100000))
Looks the same here, but anything computed from the data (boxplot stats, smoothers, densities) changes when points are dropped.
Setup for next bit
p <- ggplot(counties, aes(x = income, y = rent, color = region)) +
geom_point(size = 0.6) +
labs(
x = "Median household income ($ logged)",
y = "Median gross rent ($)",
color = NULL
) +
scale_x_log10()
p
Themes
Complete themes restyle everything at once:
p + theme_bw()
Try theme_minimal(), theme_bw(), theme_classic(), theme_void(), etc. and packages like ggthemes for more.
Scale your fonts to the medium
Every complete theme takes base_size to scale all the text together.
Paper figures ≈ 11, but slides need much more:
p + theme_minimal(base_size = 20)
Foreshadowing: What matters is font size relative to the figure. Save the same plot smaller and every label looks bigger.
Or size each piece separately
theme() targets one element at a time. Elements take an element_*():
p + theme_minimal(base_size = 15) +
theme(
axis.title.x = element_text(size = 20, face = "bold"),
axis.text = element_text(size = 10),
legend.text = element_text(size = 12, family = "Palatino")
)
The element_*() family
element_text(),element_line()element_rect()— backgrounds and borderselement_blank()— removes the element
p + theme_bw(base_size = 15) + theme(panel.grid = element_blank())
A complete theme resets everything before it.
Color palettes
Pick non-default colors or it looks like you didn’t try.
p + theme_minimal(base_size = 15) + scale_color_viridis_d()
viridis is colorblind-safe and printer-safe. scale_*_brewer() uses ColorBrewer. There are many packages (paletteer, MetBrewer, etc)
Or make your own
pal <- c(
"Midwest" = "#E87722", "Northeast" = "steelblue",
"South" = "firebrick", "West" = "#83876F"
)
p + theme_minimal(base_size = 15) + scale_color_manual(values = pal)
Composing figures: patchwork
A package can define +, |, and / on ggplot objects:
library(patchwork)
p <- p + theme_minimal(base_size = 15)
box <- ggplot(counties, aes(x = region, y = income)) +
geom_boxplot() +
theme_minimal(base_size = 15)
p + box
patchwork layouts
| puts plots beside each other; / stacks them:
hist_inc <- ggplot(counties, aes(x = income)) +
theme_minimal(base_size = 15) +
geom_histogram(bins = 40)
(p | box) / hist_inc
What’s wrong here?
Maps, the ggplot way
This could not be easier:
shapes_data |>
ggplot() +
geom_sf()
Maps, the ggplot way
geom_sf() is a geom like any other — aesthetics, scales, and themes all apply:
ggplot(shapes_data) +
geom_sf(aes(fill = income), linewidth = 0.05) +
scale_fill_viridis_c(
name = "Median income",
breaks = c(50000, 100000, 150000),
labels = c("$50k", "$100k", "$150k")
) +
theme_void(base_size = 15)
Multiple maps
Income dollars dwarf rent dollars, so give each variable its own percentiles, then facet:
long <- long |>
group_by(name) |>
mutate(percentile = percent_rank(value))
ggplot(long) +
geom_sf(aes(fill = percentile), linewidth = 0.02) +
scale_fill_viridis_c() +
facet_wrap(~name) +
theme_void(base_size = 15)
Lines: type = "l"
All 3,135 counties sorted by income:
plot(sort(counties$income), type = "l")
What does the rightmost tail tell you?
Histograms
A better way to visualize the distribution of one variable (binned):
hist(counties$income, breaks = 50) ## number of bins
Bar plots: counting categories
Use barplot() for categorical variables:
table(counties$region) |> barplot()
(Categories appear alphabetically by default.)
Densities: a smooth histogram
density() estimates the distribution; plot() draws the estimate:
density(counties$income, na.rm = TRUE) |> plot()
Layer by layer
Start the canvas — every county in gray:
nj <- counties[counties$state == "New Jersey", ]
plot(counties$income, counties$rent, pch = 19, cex = 0.4, col = "gray")
Layer by layer
The canvas is live. Add points() on top:
points(nj$income, nj$rent, pch = 19, col = "orange")
Layer by layer
…and abline() draw a reference line:
abline(h = median(counties$rent, na.rm = TRUE), lwd = 2, col = "steelblue")
Every New Jersey county sits above the national median rent
Always label your layers
The same trick compares distributions — start with the Northeast:
plot(density(counties$income[counties$region == "Northeast"], na.rm = TRUE),
col = "steelblue", lwd = 2, main = "Income by region"
)
Always label your layers
lines() overlays the South:
lines(density(counties$income[counties$region == "South"], na.rm = TRUE),
col = "firebrick", lwd = 2
)
We had to know ahead of time to set ylim. Otherwise we’d have to start over. A pain that ggplot2 will solve.
Always label your layers
legend() for a legend:
legend("topright",
legend = c("Northeast", "South"),
col = c("steelblue", "firebrick"), lwd = 2
)
Many plots at once: par()
par() sets graphics options for the whole device — mfrow makes a grid that fills plot by plot:
par(mfrow = c(1, 2))
hist(counties$income, main = "Income")
hist(counties$rent, main = "Rent")
Maps are just plots
plot() knows what to do with a geometry column:
st_geometry(shapes) |> plot()
What happens if you just call plot(shapes)?
State colors
plot(shapes["state"], main = NULL)
Counties to states
shapes |>
group_by(state) |>
summarise(geometry = st_union(geometry)) |>
plot(main = NULL)